HAMMER 61B/Many: Stabilization
[dragonfly.git] / sys / vfs / hammer / hammer_reblock.c
CommitLineData
bf686dbe
MD
1/*
2 * Copyright (c) 2008 The DragonFly Project. All rights reserved.
3 *
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 *
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
16 * distribution.
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
98da6d8c 34 * $DragonFly: src/sys/vfs/hammer/hammer_reblock.c,v 1.26 2008/07/11 01:22:29 dillon Exp $
bf686dbe
MD
35 */
36/*
37 * HAMMER reblocker - This code frees up fragmented physical space
38 *
39 * HAMMER only keeps track of free space on a big-block basis. A big-block
40 * containing holes can only be freed by migrating the remaining data in
41 * that big-block into a new big-block, then freeing the big-block.
42 *
43 * This function is called from an ioctl or via the hammer support thread.
44 */
45
46#include "hammer.h"
47
36f82b23 48static int hammer_reblock_helper(struct hammer_ioc_reblock *reblock,
bf686dbe
MD
49 hammer_cursor_t cursor,
50 hammer_btree_elm_t elm);
36f82b23 51static int hammer_reblock_data(struct hammer_ioc_reblock *reblock,
bf686dbe 52 hammer_cursor_t cursor, hammer_btree_elm_t elm);
2f85fa4d
MD
53static int hammer_reblock_leaf_node(struct hammer_ioc_reblock *reblock,
54 hammer_cursor_t cursor, hammer_btree_elm_t elm);
55static int hammer_reblock_int_node(struct hammer_ioc_reblock *reblock,
bf686dbe
MD
56 hammer_cursor_t cursor, hammer_btree_elm_t elm);
57
58int
36f82b23
MD
59hammer_ioc_reblock(hammer_transaction_t trans, hammer_inode_t ip,
60 struct hammer_ioc_reblock *reblock)
bf686dbe
MD
61{
62 struct hammer_cursor cursor;
63 hammer_btree_elm_t elm;
64 int error;
a7e9bef1 65 int checkspace_count;
bf686dbe 66
dd94f1b1
MD
67 if ((reblock->key_beg.localization | reblock->key_end.localization) &
68 HAMMER_LOCALIZE_PSEUDOFS_MASK) {
69 return(EINVAL);
70 }
71 if (reblock->key_beg.obj_id >= reblock->key_end.obj_id)
bf686dbe
MD
72 return(EINVAL);
73 if (reblock->free_level < 0)
74 return(EINVAL);
75
dd94f1b1
MD
76 reblock->key_cur = reblock->key_beg;
77 reblock->key_cur.localization += ip->obj_localization;
814387f6 78
a7e9bef1 79 checkspace_count = 0;
bf686dbe 80retry:
4e17f465 81 error = hammer_init_cursor(trans, &cursor, NULL, NULL);
bf686dbe
MD
82 if (error) {
83 hammer_done_cursor(&cursor);
dd94f1b1 84 goto failed;
bf686dbe 85 }
dd94f1b1
MD
86 cursor.key_beg.localization = reblock->key_cur.localization;
87 cursor.key_beg.obj_id = reblock->key_cur.obj_id;
bf686dbe
MD
88 cursor.key_beg.key = HAMMER_MIN_KEY;
89 cursor.key_beg.create_tid = 1;
90 cursor.key_beg.delete_tid = 0;
91 cursor.key_beg.rec_type = HAMMER_MIN_RECTYPE;
92 cursor.key_beg.obj_type = 0;
93
dd94f1b1
MD
94 cursor.key_end.localization = reblock->key_end.localization +
95 ip->obj_localization;
96 cursor.key_end.obj_id = reblock->key_end.obj_id;
bf686dbe
MD
97 cursor.key_end.key = HAMMER_MAX_KEY;
98 cursor.key_end.create_tid = HAMMER_MAX_TID - 1;
99 cursor.key_end.delete_tid = 0;
100 cursor.key_end.rec_type = HAMMER_MAX_RECTYPE;
101 cursor.key_end.obj_type = 0;
102
103 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
9480ff55 104 cursor.flags |= HAMMER_CURSOR_BACKEND;
bf686dbe 105
2f85fa4d
MD
106 /*
107 * This flag allows the btree scan code to return internal nodes,
108 * so we can reblock them in addition to the leafs. Only specify it
109 * if we intend to reblock B-Tree nodes.
110 */
111 if (reblock->head.flags & HAMMER_IOC_DO_BTREE)
112 cursor.flags |= HAMMER_CURSOR_REBLOCKING;
113
bf686dbe
MD
114 error = hammer_btree_first(&cursor);
115 while (error == 0) {
2f85fa4d
MD
116 /*
117 * Internal or Leaf node
118 */
bf686dbe 119 elm = &cursor.node->ondisk->elms[cursor.index];
dd94f1b1
MD
120 reblock->key_cur.obj_id = elm->base.obj_id;
121 reblock->key_cur.localization = elm->base.localization;
bf686dbe 122
9f5097dc
MD
123 /*
124 * Yield to more important tasks
125 */
126 if ((error = hammer_signal_check(trans->hmp)) != 0)
127 break;
a7e9bef1
MD
128
129 /*
130 * If we build up too much meta-data we have to wait for
131 * a flush cycle.
132 */
06ad81ff
MD
133 if (hammer_flusher_meta_limit(trans->hmp) ||
134 hammer_flusher_undo_exhausted(trans, 2)) {
135 error = EWOULDBLOCK;
136 break;
9f5097dc
MD
137 }
138
a7e9bef1
MD
139 /*
140 * If there is insufficient free space it may be due to
141 * reserved bigblocks, which flushing might fix.
142 */
143 if (hammer_checkspace(trans->hmp, HAMMER_CHECKSPACE_SLOP_REBLOCK)) {
144 if (++checkspace_count == 10) {
145 error = ENOSPC;
146 } else {
147 error = EWOULDBLOCK;
148 }
149 break;
150 }
151
9480ff55
MD
152 /*
153 * Acquiring the sync_lock prevents the operation from
154 * crossing a synchronization boundary.
09ac686b
MD
155 *
156 * NOTE: cursor.node may have changed on return.
9480ff55 157 */
2f85fa4d 158 hammer_sync_lock_sh(trans);
36f82b23 159 error = hammer_reblock_helper(reblock, &cursor, elm);
2f85fa4d 160 hammer_sync_unlock(trans);
bf686dbe
MD
161 if (error == 0) {
162 cursor.flags |= HAMMER_CURSOR_ATEDISK;
163 error = hammer_btree_iterate(&cursor);
164 }
a7e9bef1 165
bf686dbe
MD
166 }
167 if (error == ENOENT)
168 error = 0;
169 hammer_done_cursor(&cursor);
06ad81ff
MD
170 if (error == EWOULDBLOCK) {
171 hammer_flusher_sync(trans->hmp);
172 goto retry;
173 }
bf686dbe
MD
174 if (error == EDEADLK)
175 goto retry;
19619882
MD
176 if (error == EINTR) {
177 reblock->head.flags |= HAMMER_IOC_HEAD_INTR;
178 error = 0;
179 }
dd94f1b1
MD
180failed:
181 reblock->key_cur.localization &= HAMMER_LOCALIZE_MASK;
bf686dbe
MD
182 return(error);
183}
184
185/*
186 * Reblock the B-Tree (leaf) node, record, and/or data if necessary.
187 *
9480ff55
MD
188 * XXX We have no visibility into internal B-Tree nodes at the moment,
189 * only leaf nodes.
bf686dbe
MD
190 */
191static int
36f82b23 192hammer_reblock_helper(struct hammer_ioc_reblock *reblock,
bf686dbe
MD
193 hammer_cursor_t cursor, hammer_btree_elm_t elm)
194{
43c665ae 195 hammer_mount_t hmp;
bf686dbe
MD
196 hammer_off_t tmp_offset;
197 int error;
bf686dbe
MD
198 int bytes;
199 int cur;
bf3b416b 200 int iocflags;
bf686dbe 201
bf686dbe 202 error = 0;
43c665ae 203 hmp = cursor->trans->hmp;
bf686dbe
MD
204
205 /*
206 * Reblock data. Note that data embedded in a record is reblocked
2f85fa4d
MD
207 * by the record reblock code. Data processing only occurs at leaf
208 * nodes and for RECORD element types.
bf686dbe 209 */
2f85fa4d
MD
210 if (cursor->node->ondisk->type != HAMMER_BTREE_TYPE_LEAF)
211 goto skip;
212 if (elm->leaf.base.btype != HAMMER_BTREE_TYPE_RECORD)
213 return(0);
bf686dbe 214 tmp_offset = elm->leaf.data_offset;
bf3b416b
MD
215 if (tmp_offset == 0)
216 goto skip;
217 if (error)
218 goto skip;
219
220 /*
221 * NOTE: Localization restrictions may also have been set-up, we can't
222 * just set the match flags willy-nilly here.
223 */
224 switch(elm->leaf.base.rec_type) {
225 case HAMMER_RECTYPE_INODE:
bf3b416b
MD
226 iocflags = HAMMER_IOC_DO_INODES;
227 break;
228 case HAMMER_RECTYPE_EXT:
229 case HAMMER_RECTYPE_FIX:
ea434b6f 230 case HAMMER_RECTYPE_PFS:
bf3b416b
MD
231 case HAMMER_RECTYPE_DIRENTRY:
232 iocflags = HAMMER_IOC_DO_DIRS;
233 break;
234 case HAMMER_RECTYPE_DATA:
235 case HAMMER_RECTYPE_DB:
236 iocflags = HAMMER_IOC_DO_DATA;
237 break;
238 default:
239 iocflags = 0;
240 break;
241 }
242 if (reblock->head.flags & iocflags) {
bf686dbe
MD
243 ++reblock->data_count;
244 reblock->data_byte_count += elm->leaf.data_len;
43c665ae 245 bytes = hammer_blockmap_getfree(hmp, tmp_offset, &cur, &error);
2f85fa4d
MD
246 if (hammer_debug_general & 0x4000)
247 kprintf("D %6d/%d\n", bytes, reblock->free_level);
bf3b416b
MD
248 if (error == 0 && (cur == 0 || reblock->free_level == 0) &&
249 bytes >= reblock->free_level) {
43c665ae 250 hammer_io_direct_uncache(hmp, &elm->leaf);
bf686dbe
MD
251 error = hammer_cursor_upgrade(cursor);
252 if (error == 0) {
36f82b23 253 error = hammer_reblock_data(reblock,
bf686dbe
MD
254 cursor, elm);
255 }
256 if (error == 0) {
257 ++reblock->data_moves;
258 reblock->data_byte_moves += elm->leaf.data_len;
259 }
260 }
261 }
262
2f85fa4d 263skip:
bf686dbe 264 /*
2f85fa4d 265 * Reblock a B-Tree internal or leaf node.
bf686dbe
MD
266 */
267 tmp_offset = cursor->node->node_offset;
bf3b416b 268 if (cursor->index == 0 &&
814387f6 269 error == 0 && (reblock->head.flags & HAMMER_IOC_DO_BTREE)) {
bf686dbe 270 ++reblock->btree_count;
43c665ae 271 bytes = hammer_blockmap_getfree(hmp, tmp_offset, &cur, &error);
2f85fa4d
MD
272 if (hammer_debug_general & 0x4000)
273 kprintf("B %6d/%d\n", bytes, reblock->free_level);
bf3b416b
MD
274 if (error == 0 && (cur == 0 || reblock->free_level == 0) &&
275 bytes >= reblock->free_level) {
bf686dbe
MD
276 error = hammer_cursor_upgrade(cursor);
277 if (error == 0) {
278 if (cursor->parent)
279 elm = &cursor->parent->ondisk->elms[cursor->parent_index];
280 else
281 elm = NULL;
2f85fa4d
MD
282 switch(cursor->node->ondisk->type) {
283 case HAMMER_BTREE_TYPE_LEAF:
284 error = hammer_reblock_leaf_node(
285 reblock, cursor, elm);
286 break;
287 case HAMMER_BTREE_TYPE_INTERNAL:
288 error = hammer_reblock_int_node(
289 reblock, cursor, elm);
290 break;
291 default:
292 panic("Illegal B-Tree node type");
293 }
bf686dbe
MD
294 }
295 if (error == 0) {
296 ++reblock->btree_moves;
297 }
298 }
299 }
300
301 hammer_cursor_downgrade(cursor);
302 return(error);
303}
304
305/*
306 * Reblock a record's data. Both the B-Tree element and record pointers
307 * to the data must be adjusted.
308 */
309static int
36f82b23 310hammer_reblock_data(struct hammer_ioc_reblock *reblock,
bf686dbe
MD
311 hammer_cursor_t cursor, hammer_btree_elm_t elm)
312{
313 struct hammer_buffer *data_buffer = NULL;
314 hammer_off_t ndata_offset;
315 int error;
316 void *ndata;
317
318 error = hammer_btree_extract(cursor, HAMMER_CURSOR_GET_DATA |
11ad5ade 319 HAMMER_CURSOR_GET_LEAF);
bf686dbe
MD
320 if (error)
321 return (error);
36f82b23 322 ndata = hammer_alloc_data(cursor->trans, elm->leaf.data_len,
bf3b416b 323 elm->leaf.base.rec_type,
36f82b23 324 &ndata_offset, &data_buffer, &error);
bf686dbe
MD
325 if (error)
326 goto done;
327
328 /*
329 * Move the data
330 */
10a5d1ba 331 hammer_modify_buffer(cursor->trans, data_buffer, NULL, 0);
bf686dbe 332 bcopy(cursor->data, ndata, elm->leaf.data_len);
10a5d1ba 333 hammer_modify_buffer_done(data_buffer);
bf686dbe 334
36f82b23
MD
335 hammer_blockmap_free(cursor->trans,
336 elm->leaf.data_offset, elm->leaf.data_len);
bf686dbe 337
10a5d1ba
MD
338 hammer_modify_node(cursor->trans, cursor->node,
339 &elm->leaf.data_offset, sizeof(hammer_off_t));
bf686dbe 340 elm->leaf.data_offset = ndata_offset;
10a5d1ba 341 hammer_modify_node_done(cursor->node);
bf686dbe
MD
342
343done:
344 if (data_buffer)
345 hammer_rel_buffer(data_buffer, 0);
346 return (error);
347}
348
bf686dbe 349/*
2f85fa4d
MD
350 * Reblock a B-Tree leaf node. The parent must be adjusted to point to
351 * the new copy of the leaf node.
bf686dbe 352 *
2f85fa4d 353 * elm is a pointer to the parent element pointing at cursor.node.
bf686dbe
MD
354 */
355static int
2f85fa4d
MD
356hammer_reblock_leaf_node(struct hammer_ioc_reblock *reblock,
357 hammer_cursor_t cursor, hammer_btree_elm_t elm)
bf686dbe
MD
358{
359 hammer_node_t onode;
360 hammer_node_t nnode;
361 int error;
362
363 onode = cursor->node;
36f82b23 364 nnode = hammer_alloc_btree(cursor->trans, &error);
8d0efe43 365
bf686dbe
MD
366 if (nnode == NULL)
367 return (error);
368
369 /*
370 * Move the node
371 */
09ac686b
MD
372 hammer_lock_ex(&nnode->lock);
373 hammer_modify_node_noundo(cursor->trans, nnode);
bf686dbe
MD
374 bcopy(onode->ondisk, nnode->ondisk, sizeof(*nnode->ondisk));
375
376 if (elm) {
377 /*
378 * We are not the root of the B-Tree
379 */
36f82b23 380 hammer_modify_node(cursor->trans, cursor->parent,
bf686dbe
MD
381 &elm->internal.subtree_offset,
382 sizeof(elm->internal.subtree_offset));
383 elm->internal.subtree_offset = nnode->node_offset;
10a5d1ba 384 hammer_modify_node_done(cursor->parent);
bf686dbe
MD
385 } else {
386 /*
387 * We are the root of the B-Tree
388 */
389 hammer_volume_t volume;
390
36f82b23 391 volume = hammer_get_root_volume(cursor->trans->hmp, &error);
bf686dbe
MD
392 KKASSERT(error == 0);
393
e8599db1
MD
394 hammer_modify_volume_field(cursor->trans, volume,
395 vol0_btree_root);
bf686dbe 396 volume->ondisk->vol0_btree_root = nnode->node_offset;
10a5d1ba 397 hammer_modify_volume_done(volume);
bf686dbe
MD
398 hammer_rel_volume(volume, 0);
399 }
400
b3bad96f 401 hammer_cursor_replaced_node(onode, nnode);
36f82b23 402 hammer_delete_node(cursor->trans, onode);
bf686dbe 403
b58c6388 404 if (hammer_debug_general & 0x4000) {
2f85fa4d
MD
405 kprintf("REBLOCK LNODE %016llx -> %016llx\n",
406 onode->node_offset, nnode->node_offset);
407 }
408 hammer_modify_node_done(nnode);
409 cursor->node = nnode;
410
411 hammer_unlock(&onode->lock);
412 hammer_rel_node(onode);
413
414 return (error);
415}
416
417/*
418 * Reblock a B-Tree internal node. The parent must be adjusted to point to
419 * the new copy of the internal node, and the node's children's parent
420 * pointers must also be adjusted to point to the new copy.
421 *
422 * elm is a pointer to the parent element pointing at cursor.node.
423 */
424static int
425hammer_reblock_int_node(struct hammer_ioc_reblock *reblock,
426 hammer_cursor_t cursor, hammer_btree_elm_t elm)
427{
428 hammer_node_locklist_t locklist = NULL;
429 hammer_node_t onode;
430 hammer_node_t nnode;
431 int error;
432 int i;
433
434 error = hammer_btree_lock_children(cursor, &locklist);
435 if (error)
436 goto done;
437
438 onode = cursor->node;
439 nnode = hammer_alloc_btree(cursor->trans, &error);
440
441 if (nnode == NULL)
442 goto done;
443
444 /*
445 * Move the node. Adjust the parent's pointer to us first.
446 */
447 hammer_lock_ex(&nnode->lock);
448 hammer_modify_node_noundo(cursor->trans, nnode);
449 bcopy(onode->ondisk, nnode->ondisk, sizeof(*nnode->ondisk));
450
451 if (elm) {
452 /*
453 * We are not the root of the B-Tree
454 */
455 hammer_modify_node(cursor->trans, cursor->parent,
456 &elm->internal.subtree_offset,
457 sizeof(elm->internal.subtree_offset));
458 elm->internal.subtree_offset = nnode->node_offset;
459 hammer_modify_node_done(cursor->parent);
460 } else {
461 /*
462 * We are the root of the B-Tree
463 */
464 hammer_volume_t volume;
465
466 volume = hammer_get_root_volume(cursor->trans->hmp, &error);
467 KKASSERT(error == 0);
468
469 hammer_modify_volume_field(cursor->trans, volume,
470 vol0_btree_root);
471 volume->ondisk->vol0_btree_root = nnode->node_offset;
472 hammer_modify_volume_done(volume);
473 hammer_rel_volume(volume, 0);
474 }
475
476 /*
477 * Now adjust our children's pointers to us.
478 */
479 for (i = 0; i < nnode->ondisk->count; ++i) {
480 elm = &nnode->ondisk->elms[i];
481 error = btree_set_parent(cursor->trans, nnode, elm);
482 if (error)
483 panic("reblock internal node: fixup problem");
484 }
485
486 /*
487 * Clean up.
488 *
489 * The new node replaces the current node in the cursor. The cursor
490 * expects it to be locked so leave it locked. Discard onode.
491 */
b3bad96f 492 hammer_cursor_replaced_node(onode, nnode);
2f85fa4d
MD
493 hammer_delete_node(cursor->trans, onode);
494
495 if (hammer_debug_general & 0x4000) {
496 kprintf("REBLOCK INODE %016llx -> %016llx\n",
b58c6388
MD
497 onode->node_offset, nnode->node_offset);
498 }
8d0efe43 499 hammer_modify_node_done(nnode);
bf686dbe 500 cursor->node = nnode;
09ac686b
MD
501
502 hammer_unlock(&onode->lock);
bf686dbe
MD
503 hammer_rel_node(onode);
504
2f85fa4d
MD
505done:
506 hammer_btree_unlock_children(&locklist);
bf686dbe
MD
507 return (error);
508}
509